Spiking neural networks fine-tuning for brain image segmentation.

Introduction: The field of machine learning has undergone a significant transformation with the progress of deep artificial neural networks (ANNs) and the growing accessibility of annotated data. ANNs usually require substantial power and memory usage to achieve optimal performance. Spiking neural networks (SNNs) have recently emerged as a low-power alternative to ANNs due to their sparsity nature. Despite their energy efficiency, SNNs are generally more difficult to be trained than ANNs. Methods: In this study, we propose a novel three-stage SNN training scheme designed specifically for segmenting human hippocampi from magnetic resonance images. Our training pipeline starts with optimizing an ANN to its maximum capacity, then employs a quick ANN-SNN conversion to initialize the corresponding spiking network. This is followed by spike-based backpropagation to fine-tune the converted SNN. In order to understand the reason behind performance decline in the converted SNNs, we conduct a set of experiments to investigate the output scaling issue. Furthermore, we explore the impact of binary and ternary representations in SNN networks and conduct an empirical evaluation of their performance through image classification and segmentation tasks. Results and discussion: By employing our hybrid training scheme, we observe significant advantages over both ANN-SNN conversion and direct SNN training solutions in terms of segmentation accuracy and training efficiency. Experimental results demonstrate the effectiveness of our model in achieving our design goals.

Zero-shot visual reasoning through probabilistic analogical mapping.

Human reasoning is grounded in an ability to identify highly abstract commonalities governing superficially dissimilar visual inputs. Recent efforts to develop algorithms with this capacity have largely focused on approaches that require extensive direct training on visual reasoning tasks, and yield limited generalization to problems with novel content. In contrast, a long tradition of research in cognitive science has focused on elucidating the computational principles underlying human analogical reasoning; however, this work has generally relied on manually constructed representations. Here we present visiPAM (visual Probabilistic Analogical Mapping), a model of visual reasoning that synthesizes these two approaches. VisiPAM employs learned representations derived directly from naturalistic visual inputs, coupled with a similarity-based mapping operation derived from cognitive theories of human reasoning. We show that without any direct training, visiPAM outperforms a state-of-the-art deep learning model on an analogical mapping task. In addition, visiPAM closely matches the pattern of human performance on a novel task involving mapping of 3D objects across disparate categories.

Inappropriate Testing of Streptococcal Pharyngitis in Children Aged Below 3 Years: Application of Statistical Process Control.

Group A strep tests in patients aged below 3 years are not recommended unless the patient has appropriate symptoms and a positive contact or signs of complications. The purpose of this quality improvement project was to increase the percentage of appropriately ordered strep tests among providers. Data were collected retrospectively and prospectively from 1163 patient visits. Providers were exposed to educational interventions, an electronic medical record order change, and provider feedback. Proportional control charts characterized the providers' behaviors and determined significant improvement among testing. The result was an increase in appropriate tests (13.7% to 37.8%), and the control charts showed sustainable results over time. This project demonstrates the efficacy of these methods to encourage antibiotic stewardship among providers. Furthermore, the interventions used here can be applied to other areas with low-value diagnostic testing. Future studies should investigate whether parental anxiety and educational programs influence testing and evaluate the efficacy of certain strategies.

The protective effects of miR-210 modified endothelial progenitor cells released exosomes in hypoxia/reoxygenation injured neurons.

We have previously demonstrated that endothelial progenitor cells (EPCs) provide beneficial effects on ischemic stroke by reducing oxidative stress, which could be through EPCs-released exosomes (EPC-EXs). EXs are emerging as a bioagent for mediating cell-cell communications via their carried microRNAs (miR). miR-210 is shown to provide a neuroprotection effect against ischemic stroke. Here, we aimed to determine whether the combination of EPC-EXs and miR-210 would provide an enhanced protective effect on neurons. The hypoxia and reoxygenation (H/R) model were applied to neurons to mimic the ischemic injury of neurons. EPCs were transfected with miR-210 mimic to elevate the level of miR-210 in cells and EPC-EXs (miR210-EPC-EXs). For functional studies, EPC-EXs were co-incubated with H/R-injured neurons, then the cell viability and reactive oxygen species (ROS) production were determined. The results showed 1) H/R induced apoptosis and ROS overproduction in neurons; 2) miR-210 mimic increased the level of miR-210 in both EPCs and EPC-EXs; 3) EPCs cultured in serum-free medium released more exosomes in comparison with cells grown in complete growth media, suggesting serum starving induce the release of EXs; 4) After transfection, EPCs grown in complete media had almost 50 times higher miR-210 level than EPCs had in serum-free media, while the EPCs-EXs isolated from the complete media has lower miR-210 expression than from the serum-free media in a time-dependent manner, suggesting the transfer of miR-210 through EXs; 5) After co-incubation, EPC-EXs and miR210-EPC-EXs were uptaken by neurons, and the miR-210 level in neurons was elevated by miR210-EPC-EXs; 6) miR210-EPC-EXs were more effective in promoting cell viability and decreasing apoptosis and ROS production than EPC-EXs. The present study demonstrated that EPCs-carried miR-210 could be released and transferred to neurons in a time-dependent manner and that miR-210 loading can enhance the protective effects of EPC-EXs on H/R-induced neuron apoptosis, oxidative stress, and decreased viability.

Lean Neural Networks for Autonomous Radar Waveform Design.

In recent years, neural networks have exploded in popularity, revolutionizing the domains of computer vision, natural language processing, and autonomous systems. This is due to neural networks ability to approximate complex non-linear functions. Despite their effectiveness, they generally require large labeled data sets and considerable processing power for both training and prediction. Some of these bottlenecks have been mitigated by recent increased availability of high-quality data sets, improvements in neural network development software, and greater hardware support. Due to algorithmic bloat, neural network inference times and imprecision make them undesirable for some problems where fast classical algorithm solutions already exist, other classes of algorithms, such as convex optimization, with non-trivial execution times could be reduced using neural solutions. These algorithms could be replaced with light-weight neural networks, benefiting from their high degree of parallelization and high accuracy when properly trained. Previous work has explored how low size, weight, and power (low SWaP) neural networks and neuromorphic computing can be used to improve autonomous radar waveform design techniques that currently rely on convex optimization. Autonomous radar waveform design helps meet the need for interference mitigation caused by an ever-growing number of consumer and commercial technologies which pollute the radio frequency (RF) spectrum. Spectral notching, a radar waveform design technique, augments transmitted radar waveforms to avoid frequencies with excessive interference while maintaining the integrity of the waveform. In this paper, we extend that work, demonstrating that lean neural networks and specialized hardware can improve inference time for waveform design without sacrificing accuracy. Our lean neural solution incorporates problem-specific information into the layer structures and loss functions to decrease network size and improve accuracy. We provide model outcomes implemented on radio frequency system on a chip (RFSoC) hardware that support our simulation results. Our neural network solution decreases inference time on traditional CPU hardware by 1057× and on GPU hardware accelerators by 883× while maintaining 99% cosine similarity.

Inter- and Intra-physician variation in quantifying actinic keratosis skin photodamage.

We investigated the variations in physician evaluation of skin photodamage based on a published photodamage scale. Of interest is the utility of a 10-level scale ranging from none and mild photodamage to actinic keratosis (AK). The dorsal forearms of 55 adult subjects with various amounts of photodamage were considered. Each forearm was independently evaluated by 15 board-certified dermatologists according to the Global Assessment Severity Scale ranging from 0 (less severe) to 9 (the most progressed stage of skin damage). Dermatologists rated the levels of photodamage based upon the photographs in blinded fashion. Results show substantial disagreement amongst the dermatologists on the severity of photodamage. Our results indicate that ratings could be more consistent if using a scale of less levels (5-levels or 3-levels). Ultimately, clinicians can use this knowledge to provide better interpretation of inter-rater evaluations and provide more reliable assessment and frequent monitoring of high-risk populations.

Quantifying skin photodamage with spatial frequency domain imaging: statistical results.

We investigated the change in optical properties and vascular parameters to characterize skin tissue from mild photodamage to actinic keratosis (AK) with comparison to a published photodamage scale. Multi-wavelength spatial frequency domain imaging (SFDI) measurements were performed on the dorsal forearms of 55 adult subjects with various amounts of photodamage. Dermatologists rated the levels of photodamage based upon the photographs in blinded fashion to allow comparison with SFDI data. For characterization of statistical data, we used artificial neural networks. Our results indicate that optical and vascular parameters can be used to quantify photodamage and can discriminate between the stages as low, medium, and high grades, with the best performance of ∼70%, ∼76% and 80% for characterization of low- medium- and high-grade lesions, respectively. Ultimately, clinicians can use this noninvasive approach for risk assessment and frequent monitoring of high-risk populations.